Running several Claude Code / Codex agents on one Mac's simulators without them stepping on each other's devices
retained text
I've been running multiple coding agents in parallel on a React Native app, each in its own git worktree. The code side was fine. The device side was chaos: two agents driving the same simulator, Metro servers colliding on 8081, and the Mac swapping itself to death. I built simfleet (open source, MIT) to fix that. The agent-specific parts: • Attribution. The CLI reads the Claude Code or Codex session ID from the environment and sends it with every device request, so the dashboard and a menu-bar icon show which session is driving which simulator or emulator. • Claims. simfleet claim <device> "what I'm doing" holds a device for a session until it releases it or the session exits. The bundled skill (simfleet skill install) teaches agents to check simfleet agents first and never drive a device another live session has claimed. They pick a free one or boot a new slim one instead. • No stealing. Lanes (worktree + device + leased Metro port) fail rather than take someone else's device, worktree or port, and agents only stop lanes they started. • One build for many agents. Native builds are fingerprinted and single-flight, so two agents asking for the same build share one. • Cheap devices. Everything boots slimmed (\~1.3 GB per iOS simulator instead of \~3.8), so running 5+ at once is realistic. Setup: bun add -g simfleet simfleet skill install # adds the skill for Claude Code and Codex Demo (49 s): [https://entropyconquers.github.io/simfleet/#video](https://entropyconquers.github.io/simfleet/#video) Repo: [https://github.com/entropyconquers/simfleet](https://github.com/entropyconquers/simfleet) macOS only for now. I'd love feedback from anyone running multi-agent setups on mobile apps.
document fetched 2026-09-29 00:23 · canonical https://reddit.com/r/ClaudeAI/comments/1wsb1l7/running_several_claude_code_codex_agents_on_one